Summary Recent experience in applying recurrent neural networks (RNNs) to interpreting permanent downhole gauge records has highlighted the potential utility of machine learning algorithms to learn reservoir behavior from data. The power of the RNN resides in its ability to retain information in a form of memory of previous patterns and information contained in the previous behavior of phenomena being modeled. This memory plays a role of informing the decision at the present time by using what happened in the past. This property suggests the RNN as a suitable choice to model sequences of reservoir information, even when the reservoir modeler is faced with incomplete knowledge of the underlying physical system. Convolutional neural networks (CNNs) are another variant of the machine learning algorithm that have shown promise in sequence modeling domains, such as audio synthesis and machine translation. In this study, RNNs and CNNs were applied to tasks that traditionally would be modeled by a reservoir simulator. This was achieved by formulating the relationship between physical quantities of interest from subsurface reservoirs as a sequence mapping problem. In addition, the performance of a CNN layer as compared with an RNN was evaluated systematically to investigate their capabilities in a variety of tasks of interest to the reservoir engineer. Preliminary results suggest that CNNs, with specific design modifications, are as capable as RNNs in modeling sequences of information, and as reliable when making inferences to cases that have not been seen by the algorithm during training. Design details and reasons pertaining to the way these two seemingly different architectures process information and handle memory are also discussed.

No abstract

This work investigated deep learning (DL) algorithms to forecast wells' economic potential using wellhead surface measurements. The results of DL algorithms were compared to existing solutions in the literature that aim at solving the same problem. The performance of algorithms trained on multiple input representations, resulting from applying unsupervised DL feature extraction methods, was compared to algorithms trained on raw inputs. In this work, available wellhead measurements that can be collected at surface were used in modeling their relationship to well production rate. A search method was used to select the best combination of measurements for modeling. In addition, features were extracted automatically using autoencoders to generate additional inputs. The algorithms used for forecasting were designed using standard and customized DL layers. The results of both DL and existing solutions from the literature were compared systematically to determine the usefulness of the DL methodology developed. Previous work has shown the limitations of correlation-based methods in offering a reliable solution when applied to data outside of the historically observed range. In addition, multiple models are sometimes required depending on operational settings, e.g. choke opening, making the modeling process inefficient. Moreover, mass balance-based methods require information, such as phase density, that might not be readily available when modeling is performed, affecting the reliability of the solutions. We found that using DL algorithms offers a flexible solution that can help overcome these limitations and opens a new path to solve similar problems. We found that algorithms trained properly on relevant features are useful tools to use in assessing the economic potential of wells, using already available surface measurements. The study developed a novel methodology for forecasting economic potential using wellhead surface measurements and demonstrated the application to real field data. The method was found to be an efficient utilization of data that are readily available and do not require well intervention to measure.

Summary This study investigated the ability to produce accurate multiphase flow profiles simulating the response of producing reservoirs, using generative deep learning (GDL) methods. Historical production data from numerical simulators were used to train a variational autoencoder (VAE) algorithm that was then used to predict the output of new wells in unseen locations. This work describes a procedure in which data analysis techniques can be applied to existing historical production profiles to gain insight into field-level reservoir flow behavior. The procedure includes clustering, dimensionality reduction, correlation, in addition to novel interpretation methodologies that synthesize the results from reservoir simulation output, characterizing flow conditions. The insight was then used to build and select samples to train a VAE algorithm that reproduces the multiphase reservoir behavior for unseen operational conditions with high accuracy. Furthermore, using deep feature space interpolation, the trained algorithm can be used to further generate new predictions of the reservoir response under operational conditions for which we do not have previous examples in the training data set. It is found that VAE can be used as a robust multiphase flow simulator. Applying the methodology to the problem of determining multiphase production rate from new producing wells in undrilled locations showed positive results. The methodology was tested successfully in predicting multiphase production under different scenarios including multiwell channelized and heterogeneous reservoirs. Comparison with other shallow supervised algorithms demonstrated improvements realized by the proposed methodology. The study developed a novel methodology to interpret both data and GDL algorithms, geared toward improving reservoir management. The method was able to predict the performance of new wells in previously undrilled locations, potentially without using a reservoir simulator.

This study investigated the ability to produce accurate multiphase flow profiles simulating the response of producing reservoirs, using Generative Deep Learning (GDL) methods. Historical production data from numerical simulators were used to train a GDL model that was then used to predict the output of new wells in unseen locations. This work describes a procedure in which data analysis techniques are used to gain insight into reservoir flow behavior at a field level based on existing historical data. The procedure includes clustering, dimensionality reduction, correlation, in addition to novel interpretation methodologies that synthesize the results from reservoir simulation output, characterizing flow conditions. The insight was then used to build and train a GDL algorithm that reproduces the multiphase reservoir behavior for unseen operational conditions with high accuracy. The trained algorithm can be used to further generate new predictions of the reservoir response under operational conditions for which we do not have previous examples in the training data set. We found that the GDL algorithm can be used as a robust multiphase flow simulator. In addition, we showed that the physics of flow can be captured and manipulated in the GDL latent space after training to reproduce different physical effects that did not exist in the original training data set. Applying the methodology to the problem of determining multiphase production rate from new producing wells in undrilled locations showed positive results. The methodology was tested successfully in predicting multiphase production under different scenarios including multiwell channelized and heterogeneous reservoirs. Comparison with other shallow supervised algorithms demonstrated improvements realized by the proposed methodology, compared to existing methods. The study developed a novel methodology to interpret both data and GDL algorithms, geared towards improving reservoir management. The method was able to predict the performance of new wells in previously undrilled locations without using a reservoir simulator.